Project Overview
Consilium (Arabic: مجلس) is a Next.js AI chat application that provides a multi-model council-style AI assistant experience. The app allows users to have conversations with an AI "council" that generates responses based on multiple perspectives, aggregated through Gradio's Karpathy LLM Council. Users can engage in rich conversations that benefit from diverse AI viewpoints, with persistent storage and automatic conversation organization.
Problem & Solution
Problem:
Users often struggle to get comprehensive answers from single AI models, which may have blind spots or limited perspectives. Traditional chat interfaces provide linear conversations without capturing the nuanced decision-making process that comes from considering multiple viewpoints.
Solution:
Consilium addresses this by implementing a "council of experts" approach:
- Multi-Perspective Responses - Uses Gradio's Karpathy LLM Council to aggregate responses from multiple AI models
- Memory-Aware Conversations - Uses Groq's LLM to maintain a summarized memory of conversation history
- Persistent Conversations - All conversations stored locally in IndexedDB
- Auto-Organization - Conversations are automatically named based on their content
Key Features
Core AI Features
Multi-Model AI Council
Integrates with Gradio's Karpathy LLM Council to generate responses from multiple AI models, providing diverse perspectives.
Conversation Management
Create, switch, delete, and auto-rename conversations seamlessly.
Real-time Chat Interface
Rich markdown rendering with GFM support and LaTeX math using KaTeX.
IndexedDB Storage
Large capacity local storage for persistent conversation history.
Responsive Design
Desktop sidebar and mobile drawer layouts for all device sizes.
Bilingual Support
Full Arabic RTL support with CSS flexbox considerations.
Advanced Features
Two-Stage Memory
Groq LLM summarizes conversation history into memory (max 150 words).
URL-Based Sharing
Share conversations via URL-based conversation sharing.
Tech Stack
Architecture Highlights
- Two-stage memory approach: Groq LLM summarizes conversation history into memory (max 150 words) for efficient context management.
- Multi-model response parsing: Implemented response parsing logic in API to handle aggregated council outputs.
- IndexedDB storage: Large storage capacity with asynchronous operations for persistent conversations.
- URL-based sharing: Conversation sharing through URL parameters.
Challenges & Solutions
- Managing Conversation Context → Two-stage memory summarization approach using Groq LLM
- Multi-Model Response Parsing → Implemented response parsing logic in API to aggregate multiple model outputs
- Browser Storage Limitations → Used IndexedDB for large capacity storage with asynchronous operations
- Mobile Responsiveness → daisyUI Drawer component for mobile-friendly navigation
- Arabic/RTL Support → CSS flexbox with RTL considerations for bilingual support
Screenshots
Future Improvements
- User authentication and cloud sync
- Export conversations to PDF/Markdown
- Customizable AI council models
- Voice input/output support
- Plugin system for extensions